Unverified50% confidenceFactExact time
Logits are the raw unnormalized scores output by the model's final layer; after softmax they become the probability distribution over each token, and they are the most observable quantity closest to "model confidence."
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverifiedlogit函数定义为log(p/(1-p))，将概率值从[0,1]映射到(-∞,+∞)，是广义线性模型中处理二分类因变量的核心技术68% similarUnverified模型对所有Token计算原始得分Logit后，通过Softmax函数将得分归一化为概率分布，所有Token的概率之和等于165% similarUnverifiedLogistic regression assumes that the log-odds is a linear combination of features, maps to probability space via the sigmoid function, and uses cross-entropy (log-likelihood) as the objective function.65% similarUnverifiedLarge language models are fundamentally performing probabilistic predictions of the next most likely token in a sequence.64% similarUnverifiedThe core mechanism of large model inference is the continuous prediction of the probability of the next word, converting raw scores into a probability distribution summing to 1 via the Softmax function.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/912212API
curl https://kongchang.com/api/v1/knowledge/claims/912212MCP
get_claim(id=912212)